mirror of
https://gh-proxy.org/https://github.com/klinsmaya/multica.git
synced 2026-11-09 01:46:46 +08:00
* feat(plugins): skill resources into the existing skill table (MUL-6350 PR 4, 1/3) A plugin's skill resource becomes an ordinary row in the skill table. Not a plugin-owned copy, not a bundle, not an artifact with a digest: the previous plugin system built all of that across fourteen tables and, at the end of it, delivered one SKILL.md that this table could already hold. The only thing the platform has to remember is which installation contributed which skill, so uninstall removes exactly those and nothing a person wrote. That is one nullable column. The upsert is scoped to rows this installation owns, so a plugin claiming a name somebody already used fails the install instead of silently replacing their work — and the test asserts the human's content survives. Upgrade prunes what the new manifest no longer declares, or a renamed skill would leave its predecessor behind with nothing to attribute it to. Fresh install is now a transaction too. It was a single INSERT; skills have to land in the same commit, because an installation whose skills half-arrived is worse than one that failed — the missing half is invisible. Skill resources flip on here rather than with the rest of PR 4: nothing executes, so it does not wait for the daemon-side MCP server that the agent trigger and mcp transport still need. Co-authored-by: multica-agent <[email protected]> * feat(plugins): agent trigger — hooks as MCP tools (MUL-6350 PR 4, 2/3) The fourth call site, and the only one that is not the host deciding to call something: an agent reads a tool description and chooses. That choice is why hooks may be reached from an agent at all — the alternative, a hook that must run before or after every turn, is a third party holding the product's main loop open, and no such position exists anywhere in this design. The daemon renders the tools but does not call the plugin. A tool call goes back to Multica, which makes the signed request. Two reasons, and the second matters more: the signing secret is derived from the deployment key and putting it on every laptop running an agent would hand out a credential that can impersonate the server to every plugin backend; and routing through the server keeps the rate limit, circuit breaker, `net:` destination check and invocation record on one code path for all four triggers instead of a daemon-side copy that would drift. A failing hook is a TOOL error, not a transport error. The agent reads it, works around it, and finishes the issue — an unreachable plugin endpoint must not fail somebody's task. The server returns 200 with a status for the same reason: a non-2xx would look to the daemon like the broker itself is broken. TestPluginToolNameSeparatorCannotBeForged caught a real flaw in my first naming scheme. Sanitizing both halves and joining with an underscore is not injective — a plugin key uses `.` and `-`, both of which have to become `_`, so `a.b` and `a-b` collapse together and `a.b_` + `c` collides with `a.b` + `c`. Two plugins would have been offering the agent one tool name. The prefix now carries a short digest of the full key rather than a lossy transliteration. Co-authored-by: multica-agent <[email protected]> * feat(plugins): mcp transport with pinned tool approval (MUL-6350 PR 4, 3/3) An `mcp` hook adopts tools from an MCP server the plugin author already runs. The difference from an `http` hook is who decides the shape: an http hook declares one endpoint in a manifest an administrator read, while an MCP server decides its own tool list at runtime. So installing the plugin is not the grant — an administrator discovers the tools and pins a specific set by name and schema digest, and a tool that appears later is not adopted until approved. Approved hooks reach the daemon as ordinary remotemcp.Connection values, so the existing broker proxies them and validatePinnedRemoteMCPTools already refuses to start when an approved tool went missing or its schema drifted. No new enforcement path. Plugin connections carry a `plugin:` contribution-id prefix. They share the broker and the resolver with a workspace's own Remote MCP connections but keep their credential in the Plugin's secret storage, which a different route serves, and the contribution id is all the broker hands back at dial time. Flips TransportMCP on in HostCapabilities, completing PR 4's capability set. Co-authored-by: multica-agent <[email protected]> * test(plugins): cover mcp connection startup via the dev-origin seam The `http` transport has had MULTICA_PLUGIN_DEV_ORIGINS since the hook engine landed: an operator names exact origins, and endpoints there skip the public-internet requirement so a plugin author can develop against a server on their own machine. The `mcp` transport never got it. Two symptoms, one cause. A plugin author could not point an mcp hook at a local MCP server, and — for the same reason — the daemon's connection startup could not be reached by a test, because it validates the endpoint as public HTTPS and the repo's MCP fixture serves plain HTTP on loopback. That left the layer carrying the entire approval guarantee at 0% coverage: startTaskRemoteMCPBrokers 0.0% -> 59.5% validatePinnedRemoteMCPTools 0.0% -> 100.0% The seam is off unless the variable is set, and a test asserts that: the same loopback connection is refused when it is empty. It skips only the public-internet requirement — the net: scope still applies, and the dial-time host pin still refuses a redirect that changes host. Tests cover what the approval is for: an approved tool the server dropped, an approved tool whose schema drifted, and an optional connection that fails without failing the task. Each was verified by breaking the guard it covers. Co-authored-by: multica-agent <[email protected]> * feat(plugins): a real example plugin, and the two bugs it found examples/plugins/deploy-sentinel is an incident-response plugin: it correlates an issue with recent deploys, files rollback requests under the team's own rules (refusing deploys past a window, and reasons too short to be evidence), pages on-call on issue events, and adopts an external metrics MCP server. Four triggers, both transports, and a skill that tells an agent when to use any of it. The other examples each demonstrate one mechanism. This one was written to find out whether the mechanisms compose, and it found two things that no unit test was ever going to: 1. mcp discovery could not succeed against ANY server. DiscoverMCPHookTools passed a nil protocol-version list, and Discover checks the server's answer against that same slice — empty means nothing is acceptable. Discover now falls back to SupportedProtocolVersions() when the caller passes none. 2. A hook handler had no way to read its own configuration. The manifest's config block drives a host-rendered form, the Action API has no config endpoint, and the hook body did not carry the values — so a plugin would have had to keep a second copy on its own server. The body now carries non-secret config. Secret-typed fields are excluded; they live in a separate table and are the plugin's credentials for its own services. Also lets an mcp dev origin trust MULTICA_PLUGIN_DEV_CA, mirroring the http transport. The manifest validator requires HTTPS, so without this a local MCP server could not be developed against at all. plugin_example_test.go installs the real manifest from examples/, not a copy, so a broken example fails the build rather than rotting quietly. Migrations renumbered to 368/369 after main advanced. 362 is added to the legacy-duplicate allowlist: two PRs took that prefix within the same hour and both merged, and the ledger keys on the full stem, so renaming one would re-run it everywhere it has already applied. main is red on this today. Co-authored-by: multica-agent <[email protected]> * fix(plugins): the deploy-sentinel example servers must serve HTTPS Both .mjs servers used node:http. A hook's transport URL has to be an https:// URL or the manifest fails validation, so the example as shipped could not actually be installed against a running Multica — the README told you to set MULTICA_PLUGIN_DEV_ORIGINS to https:// origins that nothing was listening on. Found by running it: install, config, skill contribution and live MCP tool discovery all work against the real servers once they speak TLS. They read TLS_CERT / TLS_KEY and the README carries the openssl one-liner. MULTICA_PLUGIN_DEV_CA changes which certificate is trusted; it never disables verification, which is why a self-signed certificate is needed rather than plain HTTP. Co-authored-by: multica-agent <[email protected]> --------- Co-authored-by: Lambda <[email protected]> Co-authored-by: multica-agent <[email protected]>
481 lines
18 KiB
Go
481 lines
18 KiB
Go
package handler
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"encoding/json"
|
|
"encoding/pem"
|
|
"fmt"
|
|
"net/http"
|
|
"net/http/httptest"
|
|
"net/url"
|
|
"os"
|
|
"path/filepath"
|
|
"runtime"
|
|
"strings"
|
|
"testing"
|
|
|
|
"github.com/multica-ai/multica/server/internal/service"
|
|
"github.com/multica-ai/multica/server/internal/util/secretbox"
|
|
"github.com/multica-ai/multica/server/pkg/plugincontract"
|
|
"github.com/multica-ai/multica/server/pkg/remotemcp"
|
|
)
|
|
|
|
// examples/plugins/deploy-sentinel, driven end to end.
|
|
//
|
|
// This installs the REAL manifest from the repository, not a fixture written
|
|
// next to the assertions. That is the point: a hand-written fixture keeps
|
|
// passing after the example rots, and the example is what a plugin author
|
|
// copies. If somebody breaks the manifest, this is what says so.
|
|
//
|
|
// The plugin declares all four triggers and both transports, so one install
|
|
// covers the whole PR-4 surface: a skill resource lands in the skill table, its
|
|
// http hooks reach an agent as tools and are invoked over a real signed request,
|
|
// and its mcp hook is discovered and approved.
|
|
//
|
|
// The only thing rewritten on the way in is the two endpoint hosts — a test
|
|
// server's port is not knowable when the manifest is written. Everything else,
|
|
// including every tool description an agent reads, is the file's own text.
|
|
|
|
func exampleRepoDir(t *testing.T) string {
|
|
t.Helper()
|
|
_, thisFile, _, ok := runtime.Caller(0)
|
|
if !ok {
|
|
t.Fatal("cannot locate this test file")
|
|
}
|
|
// server/internal/handler -> repo root
|
|
dir := filepath.Join(filepath.Dir(thisFile), "..", "..", "..", "examples", "plugins", "deploy-sentinel")
|
|
if _, err := os.Stat(filepath.Join(dir, plugincontract.ManifestFilename)); err != nil {
|
|
t.Fatalf("the example plugin is missing: %v", err)
|
|
}
|
|
return dir
|
|
}
|
|
|
|
type exampleServers struct {
|
|
sentinel *httptest.Server
|
|
metrics *httptest.Server
|
|
}
|
|
|
|
func hostOf(t *testing.T, rawURL string) string {
|
|
t.Helper()
|
|
parsed, err := url.Parse(rawURL)
|
|
if err != nil {
|
|
t.Fatalf("parse %q: %v", rawURL, err)
|
|
}
|
|
return parsed.Hostname()
|
|
}
|
|
|
|
// stageExamplePlugin copies the example into a temp MULTICA_PLUGIN_DIR and
|
|
// repoints its two endpoints (and the net: scopes that authorise them) at the
|
|
// running test servers.
|
|
func stageExamplePlugin(t *testing.T, servers exampleServers) (string, []string) {
|
|
t.Helper()
|
|
source := exampleRepoDir(t)
|
|
root := t.TempDir()
|
|
staged := filepath.Join(root, "deploy-sentinel")
|
|
|
|
if err := os.CopyFS(staged, os.DirFS(source)); err != nil {
|
|
t.Fatalf("stage the example: %v", err)
|
|
}
|
|
|
|
manifestPath := filepath.Join(staged, plugincontract.ManifestFilename)
|
|
raw, err := os.ReadFile(manifestPath)
|
|
if err != nil {
|
|
t.Fatalf("read the example manifest: %v", err)
|
|
}
|
|
sentinelHost, metricsHost := hostOf(t, servers.sentinel.URL), hostOf(t, servers.metrics.URL)
|
|
rewritten := strings.NewReplacer(
|
|
"https://sentinel.example.com", servers.sentinel.URL,
|
|
"https://metrics.example.com", servers.metrics.URL,
|
|
"net:sentinel.example.com", "net:"+sentinelHost,
|
|
"net:metrics.example.com", "net:"+metricsHost,
|
|
).Replace(string(raw))
|
|
// A net: scope names a host, not a URL, and both test servers land on the
|
|
// same loopback host with different ports. Two identical scope lines are a
|
|
// manifest error, so collapse them — this is an artefact of the rewrite,
|
|
// not of the example.
|
|
rewritten = dedupeScopes(t, rewritten)
|
|
if err := os.WriteFile(manifestPath, []byte(rewritten), 0o644); err != nil {
|
|
t.Fatalf("write the staged manifest: %v", err)
|
|
}
|
|
|
|
// Read the scopes back out of the staged file rather than listing them
|
|
// here: granting a set this test maintains by hand would let the manifest
|
|
// add a scope that nothing ever notices.
|
|
var parsed struct {
|
|
Scopes []string `json:"scopes"`
|
|
}
|
|
if err := json.Unmarshal([]byte(rewritten), &parsed); err != nil {
|
|
t.Fatalf("parse the staged manifest: %v", err)
|
|
}
|
|
return root, parsed.Scopes
|
|
}
|
|
|
|
// dedupeScopes rewrites the manifest's scope array in place, dropping repeats
|
|
// while preserving order.
|
|
func dedupeScopes(t *testing.T, manifest string) string {
|
|
t.Helper()
|
|
var document map[string]json.RawMessage
|
|
if err := json.Unmarshal([]byte(manifest), &document); err != nil {
|
|
t.Fatalf("parse the staged manifest: %v", err)
|
|
}
|
|
var scopes []string
|
|
if err := json.Unmarshal(document["scopes"], &scopes); err != nil {
|
|
t.Fatalf("parse scopes: %v", err)
|
|
}
|
|
seen := map[string]bool{}
|
|
unique := make([]string, 0, len(scopes))
|
|
for _, scope := range scopes {
|
|
if seen[scope] {
|
|
continue
|
|
}
|
|
seen[scope] = true
|
|
unique = append(unique, scope)
|
|
}
|
|
encoded, err := json.Marshal(unique)
|
|
if err != nil {
|
|
t.Fatalf("encode scopes: %v", err)
|
|
}
|
|
document["scopes"] = encoded
|
|
out, err := json.Marshal(document)
|
|
if err != nil {
|
|
t.Fatalf("encode the staged manifest: %v", err)
|
|
}
|
|
return string(out)
|
|
}
|
|
|
|
func installExamplePlugin(t *testing.T, servers exampleServers) string {
|
|
t.Helper()
|
|
withPluginsV1Flag(t, testHandler, true)
|
|
root, scopes := stageExamplePlugin(t, servers)
|
|
|
|
plugins := testHandler.PluginService
|
|
previous := *plugins
|
|
plugins.LocalDir = root
|
|
plugins.Host = plugincontract.HostCapabilities()
|
|
// The two test servers run on loopback with self-signed certs, which the
|
|
// outbound guard refuses by design. DevOrigins is the operator opt-in a
|
|
// plugin author uses for exactly this, and it is scoped to these origins.
|
|
plugins.DevOrigins = []string{servers.sentinel.URL, servers.metrics.URL}
|
|
plugins.HookClient = servers.sentinel.Client()
|
|
// The manifest declares two secret config fields, so storing configuration
|
|
// needs the secretbox the deployment would normally supply.
|
|
box, err := secretbox.New(bytes.Repeat([]byte{9}, 32))
|
|
if err != nil {
|
|
t.Fatalf("secretbox.New: %v", err)
|
|
}
|
|
plugins.Secrets = box
|
|
// Signing is what makes a hook call verifiable by the plugin's own server;
|
|
// without a deployment key the engine refuses to call out at all.
|
|
plugins.DeploymentKey = bytes.Repeat([]byte{3}, 32)
|
|
plugins.Callbacks = service.NewCallbackTokens()
|
|
plugins.CallbackBaseURL = "https://multica.test/api/v1/plugin"
|
|
t.Cleanup(func() { *plugins = previous })
|
|
|
|
body, _ := json.Marshal(map[string]any{
|
|
"source_url": service.LocalSourcePrefix + "deploy-sentinel",
|
|
"granted_scopes": scopes,
|
|
})
|
|
recorder := httptest.NewRecorder()
|
|
testHandler.InstallPlugin(recorder, pluginHandlerRequest(http.MethodPost, "/plugins", body, map[string]string{"id": testWorkspaceID}))
|
|
if recorder.Code != http.StatusCreated {
|
|
t.Fatalf("install deploy-sentinel: status=%d body=%s", recorder.Code, recorder.Body.String())
|
|
}
|
|
var installed struct {
|
|
ID string `json:"id"`
|
|
}
|
|
if err := json.Unmarshal(recorder.Body.Bytes(), &installed); err != nil {
|
|
t.Fatalf("decode installation: %v", err)
|
|
}
|
|
t.Cleanup(func() {
|
|
cleanup := httptest.NewRecorder()
|
|
testHandler.UninstallPlugin(cleanup, pluginHandlerRequest(http.MethodDelete, "/plugins", nil,
|
|
map[string]string{"id": testWorkspaceID, "installationId": installed.ID}))
|
|
})
|
|
return installed.ID
|
|
}
|
|
|
|
func configureExamplePlugin(t *testing.T, installationID string) {
|
|
t.Helper()
|
|
body, _ := json.Marshal(map[string]any{"values": map[string]any{
|
|
"service_prefix": "prod-",
|
|
"rollback_window_minutes": 60,
|
|
"require_incident_label": true,
|
|
"environment": "production",
|
|
"sentinel_token": "sentinel-secret",
|
|
}})
|
|
recorder := httptest.NewRecorder()
|
|
testHandler.ConfigurePlugin(recorder, pluginHandlerRequest(http.MethodPut, "/config", body,
|
|
map[string]string{"id": testWorkspaceID, "installationId": installationID}))
|
|
if recorder.Code != http.StatusOK {
|
|
t.Fatalf("configure: status=%d body=%s", recorder.Code, recorder.Body.String())
|
|
}
|
|
}
|
|
|
|
// quietServer is the plugin's own hook endpoint: TLS, because the http
|
|
// transport's client is supplied by the test and can trust its certificate.
|
|
func quietServer(t *testing.T, handler http.HandlerFunc) *httptest.Server {
|
|
t.Helper()
|
|
server := httptest.NewTLSServer(handler)
|
|
t.Cleanup(server.Close)
|
|
return server
|
|
}
|
|
|
|
// metricsServer is the plugin author's own MCP server: HTTPS, because the
|
|
// manifest validator requires it and a dev origin does not relax that.
|
|
//
|
|
// remotemcp.Discover builds its own HTTP client, so the only way for it to
|
|
// trust this certificate is the same one a plugin author uses locally —
|
|
// MULTICA_PLUGIN_DEV_CA, pointing at a CA bundle.
|
|
func metricsServer(t *testing.T) *httptest.Server {
|
|
t.Helper()
|
|
server := httptest.NewTLSServer(http.HandlerFunc(serveMetricsMCP))
|
|
t.Cleanup(server.Close)
|
|
|
|
caPath := filepath.Join(t.TempDir(), "dev-ca.pem")
|
|
pem := pem.EncodeToMemory(&pem.Block{Type: "CERTIFICATE", Bytes: server.Certificate().Raw})
|
|
if err := os.WriteFile(caPath, pem, 0o600); err != nil {
|
|
t.Fatalf("write the dev CA: %v", err)
|
|
}
|
|
t.Setenv(remotemcp.DevCAEnv, caPath)
|
|
t.Setenv(remotemcp.DevOriginsEnv, server.URL)
|
|
return server
|
|
}
|
|
|
|
// The example's SKILL.md becomes an ordinary workspace skill. No bundle
|
|
// pipeline, no digest, no new table — that was the whole claim.
|
|
func TestExamplePluginContributesItsSkill(t *testing.T) {
|
|
servers := exampleServers{
|
|
sentinel: quietServer(t, func(http.ResponseWriter, *http.Request) {}),
|
|
metrics: metricsServer(t),
|
|
}
|
|
installationID := installExamplePlugin(t, servers)
|
|
|
|
names := skillNamesForInstallation(t, installationID)
|
|
if len(names) != 1 || names[0] != "incident-response" {
|
|
t.Fatalf("skills for the installation = %v, want [incident-response]", names)
|
|
}
|
|
|
|
var body string
|
|
if err := testPool.QueryRow(context.Background(),
|
|
`SELECT content FROM skill WHERE plugin_installation_id = $1`, installationID).Scan(&body); err != nil {
|
|
t.Fatalf("read the skill content: %v", err)
|
|
}
|
|
// Asserting on a sentence the skill actually teaches, not on its length: a
|
|
// truncated or mis-resolved entry would still have bytes in it.
|
|
if !strings.Contains(body, "Blast radius") {
|
|
t.Fatalf("the skill did not come from the example's SKILL.md; got %d bytes", len(body))
|
|
}
|
|
}
|
|
|
|
// What an agent working an incident actually sees.
|
|
func TestExamplePluginHooksReachAnAgentAsTools(t *testing.T) {
|
|
servers := exampleServers{
|
|
sentinel: quietServer(t, func(http.ResponseWriter, *http.Request) {}),
|
|
metrics: metricsServer(t),
|
|
}
|
|
installExamplePlugin(t, servers)
|
|
|
|
tools, err := testHandler.PluginService.AgentHookTools(context.Background(), parseUUID(testWorkspaceID))
|
|
if err != nil {
|
|
t.Fatalf("agent hook tools: %v", err)
|
|
}
|
|
byHook := make(map[string]service.PluginHookTool, len(tools))
|
|
for _, tool := range tools {
|
|
byHook[tool.HookKey] = tool
|
|
}
|
|
|
|
for _, hookKey := range []string{"correlate_deploys", "request_rollback"} {
|
|
if _, ok := byHook[hookKey]; !ok {
|
|
t.Fatalf("hook %q declares the agent trigger but is not offered as a tool", hookKey)
|
|
}
|
|
}
|
|
// page_oncall is event-only. An agent being able to page on-call by
|
|
// choosing to would defeat the point of the trigger being separate.
|
|
if _, offered := byHook["page_oncall"]; offered {
|
|
t.Fatal("page_oncall is event-only and must never appear in an agent's tool list")
|
|
}
|
|
|
|
rollback := byHook["request_rollback"]
|
|
// The description is the manifest's verbatim, because that text is how the
|
|
// agent decides which tool to reach for.
|
|
if !strings.Contains(rollback.Description, "does NOT roll anything back") {
|
|
t.Fatalf("the tool description must be the manifest's own text, got: %q", rollback.Description)
|
|
}
|
|
if !strings.HasPrefix(rollback.Name, "deploy_sentinel_") || !strings.HasSuffix(rollback.Name, "__request_rollback") {
|
|
t.Fatalf("tool name %q is not namespaced by plugin", rollback.Name)
|
|
}
|
|
if len(rollback.InputSchema) == 0 || !strings.Contains(string(rollback.InputSchema), "deploy_id") {
|
|
t.Fatalf("the manifest's input schema did not reach the agent: %s", rollback.InputSchema)
|
|
}
|
|
}
|
|
|
|
// The whole round trip: an agent calls the tool, Multica signs the request, the
|
|
// plugin's own server sees that signature and answers, and the answer comes
|
|
// back. Nothing is mocked except the author's business logic, which is theirs.
|
|
func TestExamplePluginAgentHookRoundTripIsSigned(t *testing.T) {
|
|
var seenSignature, seenTimestamp, seenHook, seenTrigger string
|
|
var seenConfig map[string]any
|
|
|
|
sentinel := quietServer(t, func(w http.ResponseWriter, r *http.Request) {
|
|
var payload struct {
|
|
HookKey string `json:"hook_key"`
|
|
Trigger string `json:"trigger"`
|
|
Config map[string]any `json:"config"`
|
|
}
|
|
_ = json.NewDecoder(r.Body).Decode(&payload)
|
|
seenSignature = r.Header.Get("X-Multica-Signature")
|
|
seenTimestamp = r.Header.Get("X-Multica-Timestamp")
|
|
seenHook, seenTrigger, seenConfig = payload.HookKey, payload.Trigger, payload.Config
|
|
|
|
w.Header().Set("Content-Type", "application/json")
|
|
_, _ = fmt.Fprintf(w, `{"deploy_count":1,"summary":"1 deploy to %vcheckout-api"}`, payload.Config["service_prefix"])
|
|
})
|
|
servers := exampleServers{sentinel: sentinel, metrics: metricsServer(t)}
|
|
|
|
installationID := installExamplePlugin(t, servers)
|
|
configureExamplePlugin(t, installationID)
|
|
|
|
input := json.RawMessage(`{"service":"checkout-api","window_minutes":120}`)
|
|
result, err := testHandler.PluginService.InvokeAgentHook(
|
|
context.Background(), installationID, "correlate_deploys", parseUUID(testUserID), input)
|
|
if err != nil {
|
|
t.Fatalf("agent hook invocation: %v", err)
|
|
}
|
|
|
|
if seenHook != "correlate_deploys" || seenTrigger != "agent" {
|
|
t.Fatalf("the plugin saw hook=%q trigger=%q", seenHook, seenTrigger)
|
|
}
|
|
if seenSignature == "" || seenTimestamp == "" {
|
|
t.Fatal("the request reached the plugin unsigned")
|
|
}
|
|
// The plugin's configuration travels with the call — its business rules
|
|
// depend on it, and it must not have to keep a second copy.
|
|
if seenConfig["service_prefix"] != "prod-" {
|
|
t.Fatalf("config did not reach the hook: %v", seenConfig)
|
|
}
|
|
// Structural, not behavioural: a secret-typed value is written to a
|
|
// different table and never reaches installation.Config, so this cannot
|
|
// fail today. It is here for the day somebody merges those two tables.
|
|
if _, leaked := seenConfig["sentinel_token"]; leaked {
|
|
t.Fatal("a secret config value was sent to the hook endpoint")
|
|
}
|
|
if result.Status != "ok" {
|
|
t.Fatalf("hook result status = %q, want ok", result.Status)
|
|
}
|
|
body, _ := json.Marshal(result.Output)
|
|
if !strings.Contains(string(body), "prod-checkout-api") {
|
|
t.Fatalf("the plugin's answer did not come back to the agent: %s", body)
|
|
}
|
|
}
|
|
|
|
// Discovery lists what the metrics server offers; approval is what makes any of
|
|
// it callable. That gap is the difference between this transport and http.
|
|
func TestExamplePluginMCPToolsNeedApproval(t *testing.T) {
|
|
servers := exampleServers{
|
|
sentinel: quietServer(t, func(http.ResponseWriter, *http.Request) {}),
|
|
metrics: metricsServer(t),
|
|
}
|
|
installationID := installExamplePlugin(t, servers)
|
|
|
|
recorder := httptest.NewRecorder()
|
|
testHandler.ListPluginMCPTools(recorder, pluginHandlerRequest(http.MethodGet, "/mcp/tools", nil, map[string]string{
|
|
"id": testWorkspaceID, "installationId": installationID, "hookKey": "metrics",
|
|
}))
|
|
if recorder.Code != http.StatusOK {
|
|
t.Fatalf("discover: status=%d body=%s", recorder.Code, recorder.Body.String())
|
|
}
|
|
var discovered struct {
|
|
Tools []struct {
|
|
Name string `json:"name"`
|
|
Approved bool `json:"approved"`
|
|
} `json:"tools"`
|
|
}
|
|
if err := json.Unmarshal(recorder.Body.Bytes(), &discovered); err != nil {
|
|
t.Fatalf("decode tools: %v", err)
|
|
}
|
|
if len(discovered.Tools) != 3 {
|
|
t.Fatalf("discovered %d tools, want the metrics server's 3", len(discovered.Tools))
|
|
}
|
|
for _, tool := range discovered.Tools {
|
|
if tool.Approved {
|
|
t.Fatalf("tool %q reported as approved before anybody approved it", tool.Name)
|
|
}
|
|
}
|
|
|
|
// Nothing approved yet, so an agent gets no connection at all.
|
|
connections, err := testHandler.PluginService.AgentMCPConnections(context.Background(), parseUUID(testWorkspaceID))
|
|
if err != nil {
|
|
t.Fatalf("agent mcp connections: %v", err)
|
|
}
|
|
if len(connections) != 0 {
|
|
t.Fatalf("an unapproved mcp hook produced %d connections", len(connections))
|
|
}
|
|
|
|
approve, _ := json.Marshal(map[string]any{"tools": []string{"query_timeseries", "list_alerts"}})
|
|
recorder = httptest.NewRecorder()
|
|
testHandler.ApprovePluginMCPTools(recorder, pluginHandlerRequest(http.MethodPut, "/mcp/tools", approve, map[string]string{
|
|
"id": testWorkspaceID, "installationId": installationID, "hookKey": "metrics",
|
|
}))
|
|
if recorder.Code != http.StatusOK {
|
|
t.Fatalf("approve: status=%d body=%s", recorder.Code, recorder.Body.String())
|
|
}
|
|
|
|
connections, err = testHandler.PluginService.AgentMCPConnections(context.Background(), parseUUID(testWorkspaceID))
|
|
if err != nil {
|
|
t.Fatalf("agent mcp connections after approval: %v", err)
|
|
}
|
|
if len(connections) != 1 {
|
|
t.Fatalf("got %d connections after approval, want 1", len(connections))
|
|
}
|
|
pinned := map[string]bool{}
|
|
for _, tool := range connections[0].ApprovedTools {
|
|
pinned[tool.Name] = true
|
|
if tool.SchemaDigest == "" {
|
|
t.Fatalf("approved tool %q was pinned without a schema digest", tool.Name)
|
|
}
|
|
}
|
|
// The third tool was discovered and deliberately not ticked. If it rode
|
|
// along, the approval step would be decorative.
|
|
if len(pinned) != 2 || !pinned["query_timeseries"] || !pinned["list_alerts"] {
|
|
t.Fatalf("pinned tools = %v, want exactly the two that were approved", pinned)
|
|
}
|
|
}
|
|
|
|
// serveMetricsMCP mirrors examples/plugins/deploy-sentinel/server/metrics-mcp.mjs.
|
|
// The .mjs file is what a person runs by hand; this is what CI runs, so the two
|
|
// must offer the same tools.
|
|
func serveMetricsMCP(w http.ResponseWriter, r *http.Request) {
|
|
var request struct {
|
|
ID json.RawMessage `json:"id"`
|
|
Method string `json:"method"`
|
|
}
|
|
_ = json.NewDecoder(r.Body).Decode(&request)
|
|
w.Header().Set("Content-Type", "application/json")
|
|
|
|
send := func(result any) {
|
|
_ = json.NewEncoder(w).Encode(map[string]any{"jsonrpc": "2.0", "id": request.ID, "result": result})
|
|
}
|
|
switch request.Method {
|
|
case "initialize":
|
|
send(map[string]any{
|
|
"protocolVersion": "2025-03-26",
|
|
"capabilities": map[string]any{"tools": map[string]any{}},
|
|
"serverInfo": map[string]any{"name": "deploy-sentinel-metrics", "version": "1.0.0"},
|
|
})
|
|
case "notifications/initialized":
|
|
w.WriteHeader(http.StatusAccepted)
|
|
case "tools/list":
|
|
send(map[string]any{"tools": []map[string]any{
|
|
{"name": "query_timeseries", "description": "Query a metric over a time window.",
|
|
"inputSchema": map[string]any{"type": "object", "required": []string{"metric", "window_minutes"}}},
|
|
{"name": "list_alerts", "description": "Alerts that fired in a window.",
|
|
"inputSchema": map[string]any{"type": "object"}},
|
|
{"name": "service_dependencies", "description": "Which services this one calls.",
|
|
"inputSchema": map[string]any{"type": "object", "required": []string{"service"}}},
|
|
}})
|
|
default:
|
|
send(map[string]any{})
|
|
}
|
|
}
|